T5 Podcast Summariser

The aim of this project is to demonstrate the capabilities of Google's T5 model that has been fine-tuned on Spotify's Podcasts Dataset for automatic podcast summarisation. Simply pass in a podcast's transcript as input to the summariser, and it would output a summary containing the gist of what the podcast is about.

T5 and Spotify Podcast Dataset

• To learn more about T5, read this blog post
• To learn more about the Dataset, read its paper

Usage

This package relies on the HuggingFace NLP Library to work so you would have to install it too:

pip install transformers
pip install t5-podcast-summariser

You can now use the package as follows:

from t5_podcast_summariser import Summariser
summariser =  Summariser()

transcript = """
Full Transcript of the podcast.....
"""

summary = summariser.summarise(transcript)
print(summary)    

Below is a sample summary generated on a podcast transcript available here:

This week on the podcast, we talk to Ayodeji Ogunnia (@Ayodeji_Ogunnia) about his life as a bricklayer in Baltimore. We also hear from one of the most famous people in the world, and how he came to be a successful bricklayer. This is a great story for anyone who wants to learn more about what it means to be a good bricklayer. If you like what we do, please leave us a review on Apple Podcasts! Thanks for listening!

Future Work

In the next update, I will make it possible for the top-n sentences to be extracted from the transcript before a summary is generated.